Benchmark compares GPT-5.6, Claude Opus 4.8, Grok 4.5 on 100 frontend briefs each
Three s — GPT-5.6 Sol, , and Grok 4.5 — were each given the same 100 frontend s, producing 300 websites in total. The briefs spanned unrelated categories like , deep tech, skincare, streetwear, and coffee. All results are published on a benchmark site called Sitegeist, where visitors can compare the three models' outputs on the same brief side by side, or browse a single model's work to spot recurring visual habits: typography choices, hero layout patterns, color palettes, geometric elements, information density, and overall composition.
The creator states this isn't meant to reduce design quality or originality to a single objective score; the value lies in the scale and consistency — the same 100 tasks given to three models, with all 300 outputs viewable rather than a handful of ed examples. The results can be browsed at sitegeist.kian.im.
Key points
- GPT-5.6 Sol, , and Grok 4.5 each got the same 100 frontend s, yielding 300 total websites
- Briefs covered unrelated categories: , deep tech, skincare, streetwear, coffee
- Results published on a benchmark site called Sitegeist (sitegeist.kian.im) for side-by-side model comparison and per-model pattern browsing (typography, layout, color, etc.)
- Creator frames the goal as scale and consistency of comparison, not a single objective design-quality score
- Creator discloses they built the benchmark themselves